arXiv AI Papers

CTHA: Constrained Temporal Hierarchical Architecture for Stable Multi-Agent LLM Systems

Back to overview

Researchers introduce CTHA, a framework stabilizing multi-agent LLM systems with temporal hierarchies. By constraining inter-layer communication through structured manifolds and arbitration mechanisms, CTHA reduces error cascades by 47% and improves sample efficiency 2.3x. The approach enables coherent decision-making across cognitive layers while maintaining coordination stability in complex autonomous systems.